Analysis of light field imaging and segmentation on all-focus images

Abstract

Image analysis is one of the fastest-growing research areas, and within it, light field or plenoptic image processing has gained great popularity in the recent years. Plenoptic images have got the inbuilt capability of imaging an object at slightly variant viewing angles which opens up possibilities for a deeper image analysis. A light field image is basically constituted with an array of sub-aperture images that have been captured using an array of lenses. Segmenting an image into its constituent parts of interest is very crucial in an image analysis point of view. Its efficiency influences the subsequent image processing stages. Various segmentation algorithms for conventional images have been developed over the recent years. This paper analyses the field of light field imaging in terms of its acquisition and discusses the various application areas and challenges pertaining to it. It also throws light on some of the research possibilities in this field. Finally, few segmentation algorithms are analysed on how it works when applied to light field all-focus images. The segmentation parameters like jaccard, dice, sensitivity, accuracy and specificity are being analysed here. From among the available sub-aperture images, the central view all-focus image has been chosen to evaluate the segmentation results. The segmentation is being evaluated on light field saliency dataset (LFSD) which has the ground truth data of segmentation corresponding to all-focus light field images.